These are chat archives for dropbox/pyston

25th
Jul 2015
Kevin Modzelewski
@kmod
Jul 25 2015 00:18
what do you think about the multiple-slowpath idea?
I've run into a number of cases where we'd be able to use it
latest one is in things like super(Foo, self).bar()
where we can't rewrite it since we don't know how super.tp_getattro will work out (though we could rewrite into that I guess)
Travis Hance
@tjhance
Jul 25 2015 00:19
Yah that sounds like a good idea
Kevin Modzelewski
@kmod
Jul 25 2015 00:21
interested in taking it on?
Travis Hance
@tjhance
Jul 25 2015 00:24
Yeah
Chris Toshok
@toshok
Jul 25 2015 02:53
lto1: fatal error: write: No space left on device
i really wish it would also tell you what path it was writing to
also, I’ll pay money if someone can explain what’s going on with this error:
../../src/runtime/cxx_unwind.cpp:664:27: error: type of ‘_ZTIN6pyston7ExcInfoE’ does not match original declaration [-Werror]
 extern "C" std::type_info EXCINFO_TYPE_INFO;
                           ^
../../src/jit.cpp:392:30: note: previously declared here
             } catch (ExcInfo e) {
                              ^
Kevin Modzelewski
@kmod
Jul 25 2015 03:14
how are you getting that?
that's some hacks (we have an extern "C" declaration of a mangled name) so maybe it's not horribly surprising
it's because we want to access typeid(ExcInfo) even though we have rtti turned off
Chris Toshok
@toshok
Jul 25 2015 03:17
that’s with -flto and gcc 4.9
it looks like the warning has the code sections mixed up though
first code snippet is the decl, second is the use, but the messages go 1) use, 2) decl
lto at least runs out of space in /tmp
creates giant .o’s - I saw one that was 292M
Kevin Modzelewski
@kmod
Jul 25 2015 03:56
well, I think the use of a type as an exception specifier causes an implicit declaration of its typeinfo
I wonder if gcc uses a different type internally?
Chris Toshok
@toshok
Jul 25 2015 03:56
ah I hadn’t realized that EXCINFO_TYPE_INFO was a #define for _ZTIN6pyston7ExcInfoE
hm, you mean a different mangling, or a different type altogether?
Kevin Modzelewski
@kmod
Jul 25 2015 03:58
sounds like they use the same mangling but different types
I guess they're not really under any contract to use std::type_info internally
like, I don't know how it the use of the typeinfo ends up getting represented internally for them
Chris Toshok
@toshok
Jul 25 2015 03:59
nod
i would have thought that would have bit us in other ways before now, though
Marius Wachtler
@undingen
Jul 25 2015 13:56
thanks for updating it! :-)